CN107229930A - A kind of pointer instrument numerical value intelligent identification Method and device - Google Patents
A kind of pointer instrument numerical value intelligent identification Method and device Download PDFInfo
- Publication number
- CN107229930A CN107229930A CN201710297198.8A CN201710297198A CN107229930A CN 107229930 A CN107229930 A CN 107229930A CN 201710297198 A CN201710297198 A CN 201710297198A CN 107229930 A CN107229930 A CN 107229930A
- Authority
- CN
- China
- Prior art keywords
- model
- image
- identification
- video camera
- pointer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/147—Details of sensors, e.g. sensor lenses
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/02—Recognising information on displays, dials, clocks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Vascular Medicine (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to field of machine vision, it is related to a kind of pointer instrument numerical value intelligent identification Method and device.Convolutional neural networks model is used the invention provides one kind, based on the model training method to resisting sample, training pattern carries out Intelligent Recognition to pointer instrument numerical value, and possesses recognition methods and the device of video camera shooting angle automatic adjusument technology.Device possesses shooting angle adaptation function, shooting angle can be carried out from correcting, using preceding without accurately being adjusted to it.This kind of method greatly reduces the workload of raw data acquisition using confrontation sample training convolutional neural networks identification model, identification that can simultaneously to multiple types instrument board numerical value so that need not individually be shot using multiple video cameras, save cost.The present invention has the features such as high robust, high-accuracy, recognition speed are fast, workable, transplanting facilitates.
Description
Technical field
The invention belongs to field of machine vision, it is related to a kind of pointer instrument numerical value intelligent identification Method and device, especially relates to
And a kind of recognition methods and device that can simultaneously carry out correcting certainly using confrontation sample training convolutional neural networks identification model.
Background technology
Pointer instrument is reliable and stable due to its, easy to maintenance, can react measured variation tendency, chemical industry, electric power,
Automobile and other industries have a wide range of applications.It is not only time-consuming but the identification of pointer instrument at present is completed by human eye
Laborious and inefficient, its reading is influenceed larger by subjective factor, therefore develops a set of pointer instrument numerical value automatic identification
Device is necessary.
Identification currently for pointer instrument is mostly used by binaryzation, morphological transformation, skeletal extraction etc.
The means of image procossing are pre-processed to image, then extract gauge pointer using Hough transform, and then pass through deflection angle of indicator
Calculating instrument represents number.There are problems in this kind of method, pointer straight line is carried such as other features (antidetonation oil line) on dial plate
These problems of determination that the interference that takes, pointer are turned to, the correction of tilted image, image contamination etc.；Simultaneously for pointer instrument
The automatic identification of table numerical value, people be all by video camera fixation to ensure the shooting angle of video camera, but in actual applications,
Some device systems are discontinuous operations, and instrument and the relative position of video camera can change, such as nuclear power emergency diesel-oil
Unit only can just be opened in emergency rating and routine test and used, and the pointer instrument parameter measurement of this kind equipment is not suitable for
Using fixed video camera, one side space-consuming, the identifying device most of the time is in idle state, secondly using process
In due to reasons such as overhauls of the equipments the relative position of video camera and instrument can be caused to change；Simultaneously it was noted that a lot
For pointer instrument numerical value automatic identifying method both for single instrument identification, but in actual applications, equipment
On often there are many pointer instrument parameters to be to need to read, these pointer instruments, which have, congener also has variety classes
's.
Grant number is the B of CN 102521560, and patent is entitled《High robust gauge pointer image-recognizing method》Chinese patent
The recognition methods to pointer instrument numerical value using Hough transform and central projection method is proposed, this kind of method needs artificial adjust
The position of whole video camera makes its positive alignment instrument board, and can be made in actual use due to the maintenance to equipment video camera with
The relative position of instrument changes, and will be unable to extract meter location and then identification, and this causes this kind of method operability not
By force, while this kind of method needs to select pointer central area and dial plate outline with mouse, in the field for needing real-time continuous to recognize
Do not applied in scape.
The Chinese patent of Application No. 201410855634.5《A kind of Recognition of Reading method and device of pointer instrument》
Describe a kind of method of use convolutional neural networks to pointer instrument numerical value automatic identification, this method is by dial plate to be identified
Several parts are divided into, the probability of different piece are located at using convolutional neural networks output pointer, further according to maximum probability part
Position in dial plate obtains the registration of pointer.This kind of method data during training pattern all use initial data,
One side collecting work amount is big, and secondly model generalization is indifferent；This method requires dial plate image to be identified and sample table simultaneously
Segmentation original position and segmentation direction of the disk image when carrying out image segmentation need unanimously, and which has limited its making in practice
With.
The content of the invention
The deficiency existed for existing method, uses convolutional neural networks model, based on confrontation the invention provides one kind
The model training method of sample, training pattern carries out Intelligent Recognition to pointer instrument numerical value, and possesses video camera shooting angle
The recognition methods of automatic adjusument technology and device.Identifying device and recognition methods of the present invention including pointer instrument numerical value,
Expansion narration separately below.
The set identifying device as shown in Figure 1, (5) is marked by support (1), program control head (2), video camera (3), cross positioning
With computer composition.Support (1) is placed on before dial plate to be identified (4) first, will be fixedly connected afterwards with video camera (3)
Program control head (2) is fixed on support (1), has the cross positioning mark of left and right two to make marks on meter panel, then system will be logical
Cross program control head (2) and video camera adjusts the shooting angle and zoom multiple of video camera (3) respectively, reach the default of shooting
It is required that (specific preset requirement illustrate in a specific embodiment, similarly hereinafter).
The set identifying device feature is as follows：Identifying device possesses shooting angle adaptation function, and shooting angle can be entered
Row is detected, to judge shooting angle and the change of video camera from correcting by positioning target to the cross of left and right two on meter panel
Whether times multiple is suitable, when that can not reach preset requirement, adjusts shooting angle respectively by program control head and video camera and takes the photograph
Camera zoom multiple, makes up to photographing request；Device is easily installed, and the covering device is compared to other methods, it is not necessary to image
The position of machine and instrument is completely fixed, only need to be substantially by video camera against instrument board, with dress, when that need not use
It can accept, same covering device can apply to different places.
The method of pointer instrument numerical value Intelligent Recognition provided by the present invention is using the confrontation sample training convolution generated
Neutral net, the model obtained using training carries out numerical identification, the implementation process such as institute of accompanying drawing 2 of this method to pointer instrument
Show.It specifically includes model training and Model Identification two large divisions.
The model using being trained to resisting sample and initial data, its implementation process as shown in the left-half of accompanying drawing 2,
Specifically include the collection, the generation to resisting sample, the part of the training of model four of the structure, sample of model.The structure of model is structure
The structure of whole convolutional neural networks is built, includes input layer, output layer, convolutional layer, pond layer and full articulamentum；Sample is adopted
Collection is the original image for gathering dial plate to be identified, is that next step prepares data to the generation of resisting sample and the training of model；Confrontation
The generation of sample is that the affine and perspective transform of original image progress, light and shade change, subregion that collect are cut, adds and makes an uproar
One or several processing means of sound obtain it is substantial amounts of available for model training to resisting sample；The training of model is by the original of collection
The convolutional neural networks identification model that beginning view data and the input of the confrontation sample data of generation are built in advance is trained, and is led to
The parameter of each node of toning integral mould obtains can be applied to the model of pointer instrument numerical identification.
Model Identification part flow as shown in the right half part of accompanying drawing 2, it include video camera from correction, image it is pre-
Model, the model that processing, image input are trained export the part of recognition result four.Video camera is that video camera passes through certainly from correction
Correction program adjust automatically shooting angle and zoom multiple are to reach photographing request；The pretreatment of image is by camera acquisition
RGB image to be identified is converted to gray level image, and extracts one or more of image instrument board and be marked；Image
It is that the convolutional neural networks trained before one or more instrument board images input by extraction are known to input the model trained
Other model；Model output result is that the identification model trained will be exported to images to be recognized after input images to be recognized
Recognition result.
The present invention possesses following feature compared with prior art：
(1) covering device possesses shooting angle adaptation function, shooting angle can be carried out from correcting, need not before
It is accurately adjusted.Package unit is easily installed, the mobility of this stability for improving system and device so that
Set of device can apply to several scenes, considerably increase the operability of this kind of method and the reuse of device；
(2) this kind of method is greatly reduced initial data and adopted using confrontation sample training convolutional neural networks identification model
The workload of collection, this method all has good antijamming capability to light, dial plate pollution, deflection of image etc., and experiment is proved
This kind of method has very high robustness；
(3) extraction and mark of this kind of method to many instrument boards, can identification to multiple types instrument board numerical value simultaneously, make
It need not must individually be shot using multiple video cameras, save cost；
(4) this kind of method changes the pattern that tradition individually uses CPU (Central Processing Unit) calculating, knot
The concurrent operation ability for having closed GPU (Graphic Processing Unit) high speed is handled, and greatly improves identification
Efficiency so that can apply in Real time identification.
Compared with prior art, this kind of method and device has high robust, high-accuracy, recognition speed fast, operable
Property strong, transplanting the features such as facilitate.
Brief description of the drawings
Fig. 1 is a kind of pointer instrument numerical value intelligent identification device figure；1. the program control video camera 4. of head 3. of support 2. is treated
Recognize the cross of dial plate 5. positioning mark
Fig. 2 is a kind of method flow diagram of pointer instrument numerical value Intelligent Recognition；
Fig. 3 is a kind of convolutional neural networks identification model structure chart；
Fig. 4 is pointer instrument to resisting sample product process figure；
Fig. 5 resists samples pictures for the pointer instrument of generation；
(1) instrument board without pointer that original image (2) is isolated from original image
(3) pointer isolated from original image；(4), (5), the confrontation samples pictures of (6) generation
Fig. 6 is pointer instrument numerical model identification process figure；
Fig. 7 pointer instrument numerical identification result figures；
Fig. 8 pointer instrument numerical value ONLINE RECOGNITION result figures.
Embodiment
It is an object of the invention to provide a kind of workable, recognition accuracy is high, speed is fast, with high robust
The method and device of pointer instrument numerical value Intelligent Recognition.
One big feature of the identifying device to possess shooting angle adaptation function, can be marked by prefabricated cross positioning into
Row is corrected certainly, it is not necessary to be manually adjusted.The covering device as shown in Figure 1, by support (1), program control head (2), video camera
(3), cross positioning mark (5) and computer composition.Support (1) is placed on before dial plate to be identified (4) first, afterwards will be with taking the photograph
The program control head (2) that camera (3) is fixedly connected is fixed on support (1), has the cross positioning mark of left and right two to do on meter panel
Mark, then video camera shooting image detect two crosses positioning target positions, judge cross positioning mark in the picture above and below, a left side
The position on the right side will pass through program control head with the departure degree (being indicated with pixel value) of system preliminary design position, hereafter system
(2) and video camera adjusts the shooting angle and zoom multiple of video camera (3) respectively, the degree of causing a deviation from reach requirement (preliminary design
Pixel value), this is the preset requirement of shooting.
The characteristics of pointer instrument numerical value intelligent identification Method provided by the present invention, is using generation to resisting sample
Training convolutional neural networks, the model obtained with training is identified, and it specifically includes model training and the big portion of Model Identification two
Point.
A kind of structure of the identification model as shown in Figure 3, by input layer, output layer, 5 convolution pond layer (1-5
Layer), 2 complete (6-7 layers) compositions of articulamentum.Input layer size is 128*128, correspond to the gray scale picture for instrument board to be identified
Size；First layer is convolution pond layer, and convolution kernel size is 5*5, provided with 48 convolution kernels, using the maximum ponds of 2*2, the layer
The tensor size of output is 62*62*48, and wherein 62*62 is the size of picture after convolution, and 48 be the number of convolution kernel；2nd layer extremely
4th layer is also convolution pond layer, and convolution kernel size is 3*3, and convolution kernel number is respectively 128,192,256, using the maximum ponds of 2*2
Change；5th layer is convolutional layer, and convolution kernel size is 3*3, provided with 338 convolution kernels, pondization operation is not carried out, by convolution pond
The tensor size for changing layer output is 6*6*338, and 6*6 is corresponding dimension of picture, and 338 be the number of convolution kernel；6th, 7 layers are two
Full articulamentum, neuron number is respectively 1024,512, is finally output layer, size is 200, correspond to the classification for picture
Number is planted, its numerical value represents the probable value that the picture to be identified inputted belongs to each class, that class of our select probability maximums
As recognition result, actual numerical value is then converted into according to its range.Described identification model structure is one kind therein,
Identification model structure in actual applications can adjust convolution pond according to the species of instrument board to be identified, range, accuracy of identification etc.
Change the number of plies and model parameter of layer and full articulamentum, obtain suitable identification model structure.
As shown in Figure 4, it includes following steps to the generating process to resisting sample：
Step 400, one or several and the congener instrument board original image of instrument board to be identified are gathered, by these instrument
The pointer of disk is separated, and generation does not contain the instrument board picture ((2) of such as accompanying drawing 5) of pointer and single pointer picture is (such as
(3) of accompanying drawing 5)；
Step 401, the pivot of pointer is found on the instrument board picture for not containing pointer, according to the range of instrument and
It is required that accuracy of identification determine generation samples pictures species, i.e. identification model output layer species.For example in the present embodiment,
Range is 0-10Mpa, and accuracy of identification is 0.05Mpa, then is divided into 200 classes；
Step 402, the instrument board using the pointer isolated in step 401 and without pointer, by two pictures according to rotation
Turn center superposition, during generation is per pictures, enters row stochastic affine transformation and perspective transform simultaneously to it, cut at random
The processing means such as subregion, the random light and shade change of image entirety, random addition noise spot are cut out, the instrument of 200 species is obtained
Disk image, comprising label file, by this kind of method make the picture number of each species at 60 and more than.Such as Fig. 5 (4),
(5), shown in (6) picture, for the partial agonistic samples pictures of generation；
Step 403, by original image and the confrontation samples pictures of generation and its respective label generation training identification model
Data set.
The training process of the identification model is advance described in the image data collection generated in step 400-403 is put into
The convolutional neural networks (80% data are used for training pattern, and 20% data are used for the degree of accuracy of detection model) built,
By constantly adjusting the parameter of each node in model, pre-provisioning request (degree of accuracy 70% and more than) is made up to, experiment is found
The mean error of this degree of accuracy threshold value drag identification is within 1%, and identification model is trained successfully.
As shown in Figure 6, it includes following steps to the Model Identification flow：
Step 600, device is placed on before instrument board to be identified, makes video camera substantially against meter location, in instrument board
Precalculated position (this precalculated position need to be with being consistent from cross positioning target preliminary design position in correction program) on face is fixed left
Right two crosses positioning mark；
Step 601, cross in detection image is positioned target position by activation system, video camera, passes through the water of program control head
Flat rotation, vertical rotary, the optical zoom of video camera make the position and preliminary design position of the cross positioning mark of left and right two in the picture
Departure degree (being represented with pixel value) reaches requirement, if still failing from correction adjustment by 50 times, it is impossible to reach that video camera is clapped
The preset requirement (preset requirement is above being told about) taken the photograph, then System self-test alarm；
Step 602, camera acquisition instrument board image, carries out gray processing processing to the image collected, is dropped afterwards
Dimension processing, each round meter disk in image is extracted by probability Hough circle transformation, and it enters rower according to the coordinate pair of each circle
Note, then by the image before the coordinate transformation of circular dial plate to dimension-reduction treatment, band is syncopated as on the image before dimension-reduction treatment
The image of each markd instrument board；
Step 603, the one or more instrument board image extracted in step 602 from one or more image is inputted into institute
State and calculated in the good convolutional neural networks identification model of training in advance.Particularly, (it is more than when the frequency of IMAQ is higher
10FPS), identification model is run on GPU in the step, and recognition speed can be greatly improved using concurrent operation processing；
Step 604, the output result of identification model in step 603 is converted into according to the range of dial plate to be identified actual
Reading, is then preserved according to corresponding mark in step 602, the identification of next round is carried out afterwards.
Recognition effect
It is tested, is surveyed during test 1 using 50 pointer instrument pictures using this kind of method and apparatus
Examination, as shown in Figure 7, solid line is Model Identification result to Comparative result, and dotted line is human eye reading value, and mean error is within 1%；
ONLINE RECOGNITION effect is tested during test 2, instrument uses dial thermometer, thermometer is put into hot water by room temperature, and
Afterwards while shooting video, while be identified, frame per second 30FPS, thermometer uses bimetallic thermometer, accuracy class 1.5, as a result
As shown in Figure 8, solid line is Model Identification result, and dotted line is human eye reading value.
Because this kind of thermometer precision be not high, there can be the situation of interim card, and diabatic process is slow, frequency acquisition is high,
Therefore have stepped, compared by artificial naked eyes, recognition result is basically identical, and mean error is within 1% on the graph.
Specific implementation step and structure drawing of device described above for the present invention, coordinates each figure to be explained.But this hair
It is bright to be not limited to above-described specific implementation step and structure drawing of device, it is any based on above-mentioned described for correlation implementation
The modifications or substitutions of step, it is any based on the above-mentioned described local directed complete set for related implementation steps, as long as the present invention's
In the range of realm of spirit, the present invention is belonged to.
Claims (5)
1. a kind of pointer instrument numerical value intelligent identification device, it is characterised in that including support (1), program control head (2), shooting
Machine (3), cross positioning mark (5) and computer；Support (1) is placed on before dial plate to be identified (4) first, afterwards will be with shooting
The program control head (2) that machine (3) is fixedly connected is fixed on support (1), has the cross positioning mark of left and right two to mark on meter panel
Note, then adjusts the shooting angle and zoom multiple of video camera (3) respectively by program control head (2) and video camera；
Detected by positioning target to the cross of left and right two on meter panel, to judge the shooting angle and zoom multiple of video camera
It is whether suitable, when preset requirement can not be reached, shooting angle is adjusted by program control head and video camera respectively and video camera becomes
Times multiple, makes up to photographing request；Above and below judging cross positioning mark in the picture, the position and system preliminary design position of left and right
Departure degree, hereafter system will adjust the shooting angle and zoom times of video camera (3) respectively by program control head (2) and video camera
Number, the degree of causing a deviation from reaches within requirement that departure degree is represented with pixel value, and this is the preset requirement of shooting.
2. a kind of method of pointer instrument numerical value Intelligent Recognition, including model training and Model Identification two large divisions；
It is characterized in that：The model specifically includes structure, the sample of model using being trained to resisting sample and initial data
Collection, the generation to resisting sample, the part of the training of model four；The structure of model is the knot for building whole convolutional neural networks
Structure, includes input layer, output layer, convolutional layer, pond layer and full articulamentum；The collection of sample is the original for gathering dial plate to be identified
Beginning image, is that next step prepares data to the generation of resisting sample and the training of model；To the generation of resisting sample to collecting
Original image carries out one of random affine and perspective transform, the change of random light and shade, the cutting of random partial region, random addition noise
Or it is several obtain model training to resisting sample；The training of model is the confrontation sample by the raw image data of collection and generation
The convolutional neural networks identification model that notebook data input is built in advance is trained, and is obtained by the parameter for adjusting each node of model
To the model that can be applied to pointer instrument numerical identification；
The model that from correction, the pretreatment of image, image input trains of the Model Identification part flow including video camera,
Model output recognition result four part；Video camera from correction be video camera by from correction program adjust automatically shooting angle and
Zoom multiple is to reach photographing request；The pretreatment of image is that the RGB image to be identified of camera acquisition is converted into gray scale
Image, and extract one or more of image instrument board and be marked；The model that image input is trained is to extract
The input of one or more instrument board images before the convolutional neural networks identification model that trains；Model output result is training
Good identification model is after input images to be recognized by recognition result of the output to images to be recognized.
3. method according to claim 1, it is characterised in that：A kind of structure of the identification model by input layer, output layer,
5 convolution pond layers, 2 full articulamentum compositions；Input layer size is 128*128, be correspond to as the gray scale of instrument board to be identified
The size of picture；First layer is convolution pond layer, and convolution kernel size is 5*5, provided with 48 convolution kernels, using the maximum ponds of 2*2,
The tensor size of this layer output is 62*62*48, and wherein 62*62 is the size of picture after convolution, and 48 be the number of convolution kernel；2nd
Layer to the 4th layer also be convolution pond layer, convolution kernel size be 3*3, convolution kernel number is respectively 128,192,256, using 2*2 most
Great Chiization；5th layer is convolutional layer, and convolution kernel size is 3*3, provided with 338 convolution kernels, pondization operation is not carried out, through pulleying
The tensor size of product pond layer output is 6*6*338, and 6*6 is corresponding dimension of picture, and 338 be the number of convolution kernel；6th, 7 layers are
Two full articulamentums, neuron number is respectively 1024,512, is finally output layer, correspond to the taxonomic species number for picture,
Its numerical value represents the probable value that the picture to be identified inputted belongs to each class, and we are used as knowledge at that class of select probability maximum
Other result, is then converted into actual numerical value according to its range.
4. method according to claim 1, it is characterised in that：It includes following steps：
Step 400, one or several and the congener instrument board original image of instrument board to be identified are gathered, by these instrument boards
Pointer is separated, and generation does not contain the instrument board picture and single pointer picture of pointer；
Step 401, the pivot of pointer is found on the instrument board picture for not containing pointer, according to the range of instrument and requirement
Accuracy of identification determine generation samples pictures species, i.e. identification model output layer species；
Step 402, the instrument board using the pointer isolated in step 401 and without pointer, by two pictures according in rotation
The heart is superimposed, and during generation is per pictures, is entered row stochastic affine transformation and perspective transform simultaneously to it, is cut out portion at random
The overall random light and shade of subregion, image changes, adds noise spot at random, obtains multiple types instrument board image, includes label text
Part, by this kind of method make the picture number of each species at 60 and more than；
Step 403, original image and the confrontation samples pictures of generation are trained into identification model with its respective label generation
Data set；
The training process of the identification model is that the image data collection generated in step 400-403 is put into described advance structure
Good convolutional neural networks, by constantly adjusting the parameter of each node in model, make up to the degree of accuracy more than 70%.
5. method according to claim 1, it is characterised in that include following steps：
Step 600, device is placed on before instrument board to be identified, makes video camera substantially against meter location, on meter panel
Precalculated position, this precalculated position need to be be consistent from cross positioning target preliminary design position in correction program, fixed left and right two
Individual cross positioning mark；
Step 601, cross in detection image is positioned target position by activation system, video camera, is revolved by the level of program control head
Turn, the position and preliminary design position that the optical zoom of vertical rotary, video camera makes the cross positioning of left and right two mark in the picture are deviateed
Degree reaches requirement, if still failing from correction adjustment by more than 50 times, it is impossible to reach the preset requirement that video camera is shot, then
System self-test is alarmed；
Step 602, camera acquisition instrument board image, carries out gray processing processing to the image collected, carries out afterwards at dimensionality reduction
Reason, each round meter disk in image is extracted by probability Hough circle transformation, and is marked according to the coordinate of each circle,
In the image before the coordinate transformation of circular dial plate to dimension-reduction treatment, it will be then syncopated as on the image before dimension-reduction treatment with mark
The image of each instrument board of note；
Step 603, the one or more instrument board image extracted in step 602 from one or more image is inputted described pre-
Calculated in the convolutional neural networks identification model first trained；
Step 604, the output result of identification model in step 603 is converted into actual reading according to the range of dial plate to be identified
Number, is then preserved according to corresponding mark in step 602, the identification of next round is carried out afterwards.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710297198.8A CN107229930B (en) | 2017-04-28 | 2017-04-28 | Intelligent identification method for numerical value of pointer instrument |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710297198.8A CN107229930B (en) | 2017-04-28 | 2017-04-28 | Intelligent identification method for numerical value of pointer instrument |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107229930A true CN107229930A (en) | 2017-10-03 |
| CN107229930B CN107229930B (en) | 2021-01-19 |
Family
ID=59934390
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710297198.8A Active CN107229930B (en) | 2017-04-28 | 2017-04-28 | Intelligent identification method for numerical value of pointer instrument |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107229930B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107705307A (en) * | 2017-11-09 | 2018-02-16 | 睿魔智能科技（东莞）有限公司 | A kind of shooting patterning process and system based on deep learning |
| CN107734329A (en) * | 2017-11-14 | 2018-02-23 | 浙江贝尔技术有限公司 | A kind of detecting system and detection method of baseband digital video frame per second |
| CN108182400A (en) * | 2017-12-27 | 2018-06-19 | 成都理工大学 | The recognition methods of charactron Dynamic Announce and system |
| CN108198154A (en) * | 2018-03-19 | 2018-06-22 | 中山大学 | Image de-noising method, device, equipment and storage medium |
| CN108345889A (en) * | 2018-02-27 | 2018-07-31 | 国网上海市电力公司 | A kind of application process carrying out registration identification to communication cabinet using Raspberry Pi |
| CN108491842A (en) * | 2018-03-27 | 2018-09-04 | 康体佳智能科技（深圳）有限公司 | A kind of dial plate identifying system and recognition methods based on neural network |
| CN108549840A (en) * | 2018-03-19 | 2018-09-18 | 广东核电合营有限公司 | A kind of method and device that analog instrument simultaneously and rapidly identifies |
| CN108571237A (en) * | 2018-04-20 | 2018-09-25 | 田凤香 | Fingerprint lock fingerprint angle correction method |
| CN108921203A (en) * | 2018-06-13 | 2018-11-30 | 深圳市云识科技有限公司 | A kind of detection and recognition methods of pointer-type water meter |
| CN109271989A (en) * | 2018-09-03 | 2019-01-25 | 广东电网有限责任公司东莞供电局 | A kind of hand-written test data automatic identifying method based on CNN and RNN model |
| CN110175982A (en) * | 2019-04-16 | 2019-08-27 | 浙江大学城市学院 | A kind of defect inspection method based on target detection |
| CN110363173A (en) * | 2019-07-22 | 2019-10-22 | 合肥小步智能科技有限公司 | A kind of more instrument automatic identifying methods of power distribution room crusing robot |
| CN110378257A (en) * | 2019-07-04 | 2019-10-25 | 上海图菱新能源科技有限公司 | Artificial intelligence model full process automatization system |
| CN110414510A (en) * | 2019-07-26 | 2019-11-05 | 华中科技大学 | A kind of readings of pointer type meters bearing calibration |
| CN110674889A (en) * | 2019-10-15 | 2020-01-10 | 贵州电网有限责任公司 | Image training method for ammeter terminal fault recognition |
| CN110941188A (en) * | 2018-09-25 | 2020-03-31 | 珠海格力电器股份有限公司 | Intelligent household control method and device |
| CN111666930A (en) * | 2019-03-07 | 2020-09-15 | 深圳海青智盈科技有限公司 | Meter reading method, device, system and storage medium |
| CN112347929A (en) * | 2020-11-06 | 2021-02-09 | 电子科技大学中山学院 | Pointer instrument system and monitoring method |
| CN113312941A (en) * | 2020-02-26 | 2021-08-27 | 北京同邦卓益科技有限公司 | Method and device for monitoring instrument panel |
| CN116437217A (en) * | 2023-06-13 | 2023-07-14 | 深圳太古语科技有限公司 | Shooting method of gene sequencing chip and gene sequencer |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060036967A1 (en) * | 2004-04-26 | 2006-02-16 | Crichlow Henry B | Remote meter reading using transmitted visual graphics. |
| US20080011654A1 (en) * | 2006-07-07 | 2008-01-17 | Hale Mathew S | Mail processing system with radiation filtering |
| CN102263933A (en) * | 2010-05-25 | 2011-11-30 | 杭州华三通信技术有限公司 | Intelligent monitoring method and device |
| CN104057765A (en) * | 2014-06-25 | 2014-09-24 | 深圳市新德昌金属制品有限公司 | Wine box facial tissue automatic positioning and forming machine as well as automatic positioning and adjusting method |
| CN105046320A (en) * | 2015-08-13 | 2015-11-11 | 中国人民解放军61599部队计算所 | Virtual sample generation method |
| CN105446360A (en) * | 2015-11-16 | 2016-03-30 | 深圳市神视检验有限公司 | Automatic tracking method and automatic tracking apparatus based on welding seam |
| CN105488297A (en) * | 2015-12-15 | 2016-04-13 | 东北大学 | Method for establishing complex product optimization design agent model based on small sample |
| CN105809179A (en) * | 2014-12-31 | 2016-07-27 | 中国科学院深圳先进技术研究院 | Pointer type instrument reading recognition method and device |
| CN105868756A (en) * | 2016-03-22 | 2016-08-17 | 武汉蓝焰自动化应用技术有限责任公司 | Gas meter counter direct reading system and direct reading method by applying digital image recognition technology |
| CN106296692A (en) * | 2016-08-11 | 2017-01-04 | 深圳市未来媒体技术研究院 | Image significance detection method based on antagonism network |
-
2017
- 2017-04-28 CN CN201710297198.8A patent/CN107229930B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060036967A1 (en) * | 2004-04-26 | 2006-02-16 | Crichlow Henry B | Remote meter reading using transmitted visual graphics. |
| US20080011654A1 (en) * | 2006-07-07 | 2008-01-17 | Hale Mathew S | Mail processing system with radiation filtering |
| CN102263933A (en) * | 2010-05-25 | 2011-11-30 | 杭州华三通信技术有限公司 | Intelligent monitoring method and device |
| CN104057765A (en) * | 2014-06-25 | 2014-09-24 | 深圳市新德昌金属制品有限公司 | Wine box facial tissue automatic positioning and forming machine as well as automatic positioning and adjusting method |
| CN105809179A (en) * | 2014-12-31 | 2016-07-27 | 中国科学院深圳先进技术研究院 | Pointer type instrument reading recognition method and device |
| CN105046320A (en) * | 2015-08-13 | 2015-11-11 | 中国人民解放军61599部队计算所 | Virtual sample generation method |
| CN105446360A (en) * | 2015-11-16 | 2016-03-30 | 深圳市神视检验有限公司 | Automatic tracking method and automatic tracking apparatus based on welding seam |
| CN105488297A (en) * | 2015-12-15 | 2016-04-13 | 东北大学 | Method for establishing complex product optimization design agent model based on small sample |
| CN105868756A (en) * | 2016-03-22 | 2016-08-17 | 武汉蓝焰自动化应用技术有限责任公司 | Gas meter counter direct reading system and direct reading method by applying digital image recognition technology |
| CN106296692A (en) * | 2016-08-11 | 2017-01-04 | 深圳市未来媒体技术研究院 | Image significance detection method based on antagonism network |
Non-Patent Citations (2)
| Title |
|---|
| LIMENG ZHAO ET AL.: "Research of Digital Meter Identifier Based On DSP and Neural Network", 《IST 2009 - INTERNATIONAL WORKSHOP ON IMAGING SYSTEMS AND TECHNIQUES》 * |
| 尹红敏 等: "基于Halcon的指针式仪表的读数", 《现代电子技术》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107705307A (en) * | 2017-11-09 | 2018-02-16 | 睿魔智能科技（东莞）有限公司 | A kind of shooting patterning process and system based on deep learning |
| CN107705307B (en) * | 2017-11-09 | 2021-01-01 | 睿魔智能科技（东莞）有限公司 | Shooting composition method and system based on deep learning |
| CN107734329A (en) * | 2017-11-14 | 2018-02-23 | 浙江贝尔技术有限公司 | A kind of detecting system and detection method of baseband digital video frame per second |
| CN108182400A (en) * | 2017-12-27 | 2018-06-19 | 成都理工大学 | The recognition methods of charactron Dynamic Announce and system |
| CN108345889B (en) * | 2018-02-27 | 2022-02-11 | 国网上海市电力公司 | Application method for performing reading identification on communication cabinet by utilizing raspberry party |
| CN108345889A (en) * | 2018-02-27 | 2018-07-31 | 国网上海市电力公司 | A kind of application process carrying out registration identification to communication cabinet using Raspberry Pi |
| CN108549840A (en) * | 2018-03-19 | 2018-09-18 | 广东核电合营有限公司 | A kind of method and device that analog instrument simultaneously and rapidly identifies |
| CN108198154A (en) * | 2018-03-19 | 2018-06-22 | 中山大学 | Image de-noising method, device, equipment and storage medium |
| CN108198154B (en) * | 2018-03-19 | 2020-06-26 | 中山大学 | Image denoising method, device, equipment and storage medium |
| CN108491842A (en) * | 2018-03-27 | 2018-09-04 | 康体佳智能科技（深圳）有限公司 | A kind of dial plate identifying system and recognition methods based on neural network |
| CN108571237A (en) * | 2018-04-20 | 2018-09-25 | 田凤香 | Fingerprint lock fingerprint angle correction method |
| CN108921203A (en) * | 2018-06-13 | 2018-11-30 | 深圳市云识科技有限公司 | A kind of detection and recognition methods of pointer-type water meter |
| CN109271989A (en) * | 2018-09-03 | 2019-01-25 | 广东电网有限责任公司东莞供电局 | A kind of hand-written test data automatic identifying method based on CNN and RNN model |
| CN110941188A (en) * | 2018-09-25 | 2020-03-31 | 珠海格力电器股份有限公司 | Intelligent household control method and device |
| CN111666930A (en) * | 2019-03-07 | 2020-09-15 | 深圳海青智盈科技有限公司 | Meter reading method, device, system and storage medium |
| CN110175982B (en) * | 2019-04-16 | 2021-11-02 | 浙江大学城市学院 | Defect detection method based on target detection |
| CN110175982A (en) * | 2019-04-16 | 2019-08-27 | 浙江大学城市学院 | A kind of defect inspection method based on target detection |
| CN110378257A (en) * | 2019-07-04 | 2019-10-25 | 上海图菱新能源科技有限公司 | Artificial intelligence model full process automatization system |
| CN110378257B (en) * | 2019-07-04 | 2023-12-19 | 山东巧思智能科技有限公司 | Artificial intelligent model whole process automation system |
| CN110363173A (en) * | 2019-07-22 | 2019-10-22 | 合肥小步智能科技有限公司 | A kind of more instrument automatic identifying methods of power distribution room crusing robot |
| CN110414510A (en) * | 2019-07-26 | 2019-11-05 | 华中科技大学 | A kind of readings of pointer type meters bearing calibration |
| CN110414510B (en) * | 2019-07-26 | 2021-10-08 | 华中科技大学 | Reading correction method for pointer instrument |
| CN110674889A (en) * | 2019-10-15 | 2020-01-10 | 贵州电网有限责任公司 | Image training method for ammeter terminal fault recognition |
| CN113312941A (en) * | 2020-02-26 | 2021-08-27 | 北京同邦卓益科技有限公司 | Method and device for monitoring instrument panel |
| CN112347929A (en) * | 2020-11-06 | 2021-02-09 | 电子科技大学中山学院 | Pointer instrument system and monitoring method |
| CN116437217A (en) * | 2023-06-13 | 2023-07-14 | 深圳太古语科技有限公司 | Shooting method of gene sequencing chip and gene sequencer |
| CN116437217B (en) * | 2023-06-13 | 2023-08-18 | 深圳太古语科技有限公司 | Shooting method of gene sequencing chip and gene sequencer |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107229930B (en) | 2021-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107229930A (en) | A kind of pointer instrument numerical value intelligent identification Method and device | |
| CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
| CN109948469A (en) | The automatic detection recognition method of crusing robot instrument based on deep learning | |
| CN107103298B (en) | Pull-up counting system and method based on image processing | |
| CN111160269A (en) | Face key point detection method and device | |
| CN110717903A (en) | Method for detecting crop diseases by using computer vision technology | |
| CN104835175B (en) | Object detection method in a kind of nuclear environment of view-based access control model attention mechanism | |
| CN108269250A (en) | Method and apparatus based on convolutional neural networks assessment quality of human face image | |
| CN109325395A (en) | The recognition methods of image, convolutional neural networks model training method and device | |
| CN106529559A (en) | Pointer-type circular multi-dashboard real-time reading identification method | |
| CN109117876A (en) | A kind of dense small target deteection model building method, model and detection method | |
| CN109977813A (en) | A kind of crusing robot object localization method based on deep learning frame | |
| CN110070074A (en) | A method of building pedestrian detection model | |
| CN109583324A (en) | A kind of pointer meters reading automatic identifying method based on the more box detectors of single-point | |
| CN107895160A (en) | Human face detection and tracing device and method | |
| CN107153848A (en) | Instrument image automatic identifying method based on OpenCV | |
| CN112967285B (en) | Chloasma image recognition method, system and device based on deep learning | |
| CN113159227A (en) | Acne image recognition method, system and device based on neural network | |
| CN106340000A (en) | Bone age assessment method | |
| CN108921099A (en) | Moving ship object detection method in a kind of navigation channel based on deep learning | |
| CN112613454A (en) | Electric power infrastructure construction site violation identification method and system | |
| CN107194361A (en) | Two-dimentional pose detection method and device | |
| CN110263790A (en) | A kind of power plant's ammeter character locating and recognition methods based on convolutional neural networks | |
| CN110688955A (en) | Building construction target detection method based on YOLO neural network | |
| CN107992783A (en) | Face image processing process and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |